Inspiration
I started this hackathon looking for a gap in what AI NAS can do, and most ideas I saw were about organizing files or finding photos. Meanwhile, voice-cloning "family emergency" scams keep getting more convincing. A scammer only needs a short clip of someone's voice, and what makes the call believable is everything else: a kid's school, their age, their neighborhood. That information is usually sitting in the family photos we share without thinking twice.
The phone apps that exist check one photo at a time. None of them notice that a school logo in one picture, a birthday cake in another, and a week of photos from the same street add up to a map of someone's life. A NAS already holds the whole collection, so it's the right place to look at it all together.
What it does
You point the Skill at an album, a folder, or everything you've already shared. It checks each file for clues, connects them across files, and gives you a plain report with a risk level for each family member: possible voice exposure, clear face photos, identity details, and location or routine. With your approval, it saves the report, tags the risky files, and builds a copy that's safe to share. Nothing leaves the NAS, and nothing is changed without approval.
How I built it
The Skill follows the Uliya Skill Development Standard. It hands every step to UGREEN's photo, document, file, and knowledge agents using only capabilities in the official catalog. I used Claude as a research and drafting partner along the way.
Challenges I ran into
My original plan measured how much of each person's voice appeared in shared videos. I emailed UGREEN Technical Support to confirm the transcription capability, and they told me it only accepts audio files, not video. So I redesigned the voice check: videos where a family member is recognized and there's an audio track are flagged as possible voice exposure, and audio files with real speech are confirmed. It's more honest, and it works within what the platform actually supports.
Accomplishments I'm proud of
The Skill passes the official uliya-skill-kit checker with zero errors and zero warnings. Every step is confirmed against the capability catalog, and its limits were confirmed directly with UGREEN. I'm also proud that it finds risks no single-photo tool can see, because it looks at the whole collection at once.
What I learned
Checking capabilities with the platform before building saved me from promising something the NAS can't do. Being clear about limits made the Skill better.
What's next for Impersonation Exposure Check
Tuning the vision prompts and risk levels on a real UGREEN AI NAS, and testing the full approval flow from scan to safe-to-share copy.
Built With
- local-ai
- markdown
- ugos-pro
- ugreen-ai-nas
- uliya
- uliya-skill-kit
